DO011
Physics-Guided Machine Learning and Differentiable Programming for Ocean Modelling

Session ID#: 257890

Session Description:
Ocean models are essential tools for observing the ocean, forecasting its state, and projecting future climate scenarios. In recent years, there has been a growing interest in leveraging machine learning (ML) to enhance various aspects of ocean models, including parameter calibration, subgrid-scale parameterizations, and model error correction. This has led to the emergence of a new generation of hybrid models that combine data-driven components with traditional physics-based frameworks. In parallel, differentiable programming, which has been applied in a limited number of ocean models for over 25 years, is now gaining broader traction. By leveraging automatic differentiation and gradient-based optimization, differentiable programming offers a powerful paradigm that enables the seamless integration of ML techniques into geoscientific models.

This session will explore recent advances and applications of machine learning and/or differentiable programming in ocean modelling. The ambition is to highlight cutting edge research at the intersection between oceanography, computer science and machine learning.  The session invites contributions spanning idealized to realistic contexts and covering all aspects of ocean science where these techniques are being explored or deployed. 

Index Terms:

1902 Community modeling frameworks [INFORMATICS]
1910 Data assimilation, integration and fusion [INFORMATICS]
1922 Forecasting [INFORMATICS]
1956 Numerical algorithms [INFORMATICS]
Primary Chair:  Julien LeSommer, Grenoble, FRANCE
Co-chairs:  Nora Loose, University of Bergen, Bergen, Norway, James Roland Maddison, University of Edinburgh, School of Mathematics, Edinburgh, United Kingdom and Dr. Andrea Storto, PhD, CNR Institute for Marine Science, Rome, Italy
 
Data-Driven Modeling of the Atlantic Meridional Overturning Circulation (2011620)
Balu T Nadiga1, Guang Zhao2, Zhengyu Liu3, Seungjun Lee2, Luke P Van Roekel4 and Wei Xu2, (1)Los Alamos National Laboratory, Los Alamos, NM, United States, (2)Brookhaven National Laboratory, Upton, United States, (3)The Ohio State University, Department of Geography, Columbus, United States, (4)Los Alamos National Laboratory, Los Alamos, United States
 
Physics-Informed Hybrid Neural Networks with Observational Constraints (2012966)
Youmin Tang, Hohai University, Nanjing, China
 
A Machine Learning Approach for Predicting Eddy Energy in a Scale-Aware Eddy Parameterisation and its Evaluation in an Idealised NEMO Configuration. (2013483)
Thomas Wilder1, Ranjini Swaminathan2 and Till Kuhlbrodt1, (1)National Centre for Atmospheric Science, University of Reading, Reading, United Kingdom, (2)University of Reading, Reading, United Kingdom
 
Neural Network Approach for Mesoscale Eddy Parameterization in the Unstructured Mesh Ocean Model FESOM2: Applications to Idealized Ocean Configurations (2013894)
Rajka Juhrbandt, GEOMAR Helmholtz Centre for Ocean Research Kiel, Ocean Dynamics, Kiel, Germany; Alfred Wegener Institute Helmholtz-Center for Polar and Marine Research Bremerhaven (AWI), Bremerhaven, Germany, Stephan Juricke, GEOMAR Helmholtz Centre for Ocean Research Kiel, Kiel, Germany, Thomas Jung, AWI, Bremerhaven, Germany and Peter Zaspel, Bergische Universität Wuppertal, Scientific computing and high performance computing, Wuppertal, Germany
 
Comparison of three data-driven methods for the calibration of sub-grid parametrizations (2016240)
Gabriel Mouttapa1, Julien Lesommer1, Marta Agnieszka Mrozowska2, Emmanuel Cosme1, Gregory Wagner3 and Bruno Deremble1, (1)Université Grenoble Alpes (UGA)/CNRS, Institut des Géosciences de l'Environnement (IGE), Grenoble, France, (2)Princeton University, Princeton, United States, (3)Massachusetts Institute of Technology, Cambridge, United States
 
Solitons Above the Sea: Investigating a Possible Mechanism for SST Error in Cold Currents (2019453)
Aleksey Aleynik, University of Strathclyde, Physics, Glasgow, United Kingdom and Philip Anderson, Scottish Association for Marine Science, Oban, United Kingdom
 
Online Learning in Idealized Ocean Gyres (2020709)
James Roland Maddison, University of Edinburgh, School of Mathematics, Edinburgh, United Kingdom
 
Evaluating a new hybrid sea ice model in global fully-coupled climate predictions and projections. (2020742)
William Gregory1, Mitchell Bushuk2, Yong-Fei Zhang1, Alistair Adcroft3, Laure Zanna4, Colleen McHugh5 and Liwei Jia6, (1)Princeton University, Princeton, NJ, United States, (2)NOAA/Geophysical Fluid Dynamics Laboratory, Princeton, NJ, United States, (3)Program for Atmospheric and Oceanic Sciences, Princeton University, Princeton, United States, (4)University of Oxford, Dept of Physics, Oxford, United Kingdom, (5)Science Applications International Corporation, Reston, United States, (6)NOAA/GFDL, Princeton, United States
 
Evaluating Deep-learning-based Parameterizations for Representing Submesoscale Dynamics: Insights from a Simplified Ocean Model (2022709)
Rin Irie1,2, Helen Stewart1, Masaki Hisada1 and Takaharu Yaguchi2, (1)NTT Inc., Space Environment and Energy Laboratories, Tokyo, Japan, (2)Kobe University, Graduate School of Science, Kobe, Japan
 
DJ4Oceananigans: A differentiable, Julia-based ocean general circulation model for gradient-based parameter calibration, online learning and variational data assimilation (2022794)
Mr. Joseph Kump1,2, William Moses3, Gregory Wagner4, Dhruv Apte2, Valentin Churavy5, Christopher N Hill6, Paul Berg7, Nora Loose8, Mose Giordano9, Dr. Sri Hari Krishna Narayanan, PhD10, Michel Schanen10, Sarah Williamson11, Patrick Heimbach12, Avik Pal4, Alexis Montoison13 and Alex Zinenko14, (1)UNITED STATES, (2)University of Texas at Austin, Austin, United States, (3)University of Illinois at Urbana Champaign, Urbana, United States, (4)Massachusetts Institute of Technology, Cambridge, United States, (5)Johannes Gutenberg University of Mainz, Mainz, Germany, (6)MIT, Cambridge, United States, (7)Bern University of Applied Sciences, Bern, Switzerland, (8)University of Bergen, Bergen, Norway, (9)University College London, London, United Kingdom, (10)Argonne National Laboratory, Lemont, United States, (11)University of Texas at Austin, Oden Institute, Austin, United States, (12)University of Texas at Austin, Geological Sciences, Austin, United States, (13)Argonne National Laboratory, Argonne, United States, (14)AMD, Paris, France
 
Leveraging Differentiable Programming and Online Learning for the design of Hybrid Numerical Models (2024199)
Said Ouala, IMT Atlantique, Brest, France, Etienne Meunier, INRIA, Paris, France, Ronan Fablet, IMT Atlantique, CNRS UMR Lab-STICC, Brest, France, Hugo Frezat, University Grenoble Alpes, Grenoble, France and Julien Lesommer, Université Grenoble Alpes (UGA)/CNRS, Institut des Géosciences de l'Environnement (IGE), Grenoble, France
 
Neural Synthetic Profiles from Remote Sensing and Observations (NeSPReSO) - an Open Web Interface and API (2028784)
Jose Roberto Miranda1, Olmo Zavala-Romero1, Eric Chassignet2 and Jorge Eduardo Velasco Zavala1, (1)Florida State University, Scientific Computing, Tallahassee, United States, (2)Center for Ocean-Atmospheric Prediction Studies, Florida State University, Tallahassee, FL, United States
 
Online performance of a machine learning parameterization of ocean mixed layer eddies (2029550)
Marcela Contreras1, Alexis Barge2, Julien Lesommer2, Abigail S. Bodner3 and Dhruv Balwada4, (1)IGE, Grenoble Alpes University, Grenoble, France, (2)Université Grenoble Alpes (UGA)/CNRS, Institut des Géosciences de l'Environnement (IGE), Grenoble, France, (3)Massachusetts Institute of Technology, Department of Earth, Atmospheric, and Planetary Sciences, Cambridge, United States, (4)Lamont-Doherty Earth Observatory, Columbia University, New York, United States
 
Design and implementation of a data-driven parameterization for mesoscale thickness fluxes (2029649)
Pavel Perezhogin, New York University, New York, NY, United States, Dhruv Balwada, Lamont-Doherty Earth Observatory, Columbia University, New York, United States, Alistair Adcroft, Program for Atmospheric and Oceanic Sciences, Princeton University, Princeton, United States and Laure Zanna, New York University, New York, United States
 
Calibration of a Neural Network-based Eddy Backscatter Parameterization in a Differentiable Hybrid Shallow Water Model (2030698)
Sarah Williamson1, Nora Loose2, William Moses3, Milan Kloewer4, Valentin Churavy5, Michel Schanen6, Sri Hari Krishna Narayanan6 and Patrick Heimbach7, (1)University of Texas at Austin, Oden Institute, Austin, United States, (2)[C]Worthy, LLC, Boulder, United States, (3)University of Illinois at Urbana Champaign, Urbana, United States, (4)University of Oxford, Oxford, United Kingdom, (5)Johannes Gutenberg University of Mainz, Mainz, Germany, (6)Argonne National Laboratory, Lemont, United States, (7)University of Texas at Austin, Geological Sciences, Austin, United States
 
Modified Permutation Entropy Analysis on Double-gyre Ocean Simulations (2037101)
Elnaz Naghibi, University of East London, Engineering and Construction, London, United Kingdom, Sergey Karabasov Prof, Queen Mary, University of London, School of Engineering and Materials Science, London, United Kingdom and Vasily Gryazev Dr, Brunel University, Mechanical and Aerospace Engineering, Uxbridge, United Kingdom
 
Quantifying Past Interglacial Uncertainty with the Use of Surrogate Modelling and Expert Elicitation. (2037414)
Dimitra Salmanidou, University College London, London, England, UNITED KINGDOM, Serge Guillas, University College London, Department of Statistical Science, London, United Kingdom, Matt Graham, University College London, London, United Kingdom and Lauren J Gregoire, University of Leeds, School of Earth and Environment, Leeds, United Kingdom
 
Towards Robust Deep Learning Parametrisations: Unifying Frameworks and Online Evaluation in Ocean Models (2037547)
Etienne Meunier, INRIA, Paris, France, Julie Deshayes, Institut Pierre Simon Laplace, LOCEAN, Paris, France and David Kamm, LOCEAN, Paris Cedex 05, France
 
Leveraging km-Scale Simulations for Low-Cost Data-Driven Emulation of the Global 3D Ocean State (2038293)
Janika Rhein, Nils Hutter and Florian Schuette, GEOMAR Helmholtz Centre for Ocean Research Kiel, Kiel, Germany
 
Online Learning of Stochastic Ocean Mesoscale Closures (2039940)
Martin Brolly, University of Edinburgh, Edinburgh, UNITED KINGDOM
 
Physics-inspired multiscale neural architectures for forecasting fluid and oceanic flows (2047554)
Anantha Narayanan Suresh Babu1, Akhil Sadam2, Patrick J Haley Jr1, Christopher Mirabito1 and Pierre F J Lermusiaux3, (1)Massachusetts Institute of Technology, Department of Mechanical Engineering, Cambridge, United States, (2)Massachusetts Institute of Technology, Center for Computational Science and Engineering, Cambridge, United States, (3)Massachusetts Institute of Technology, Department of Mechanical Engineering, Cambridge, MA, United States
 
Sparse and Deep Gaussian Process Closure Modelling for 2-D Fluid and Ocean Flows (2048051)
Sanaa Mouzahir, Massachusetts Institute of Technology, Mechanical engineering, Cambridge, United States and Pierre F J Lermusiaux, Massachusetts Institute of Technology, Department of Mechanical Engineering, Cambridge, MA, United States
See more of: Digital Ocean
<< Previous Session Proposal | Next Session Proposal >>